The queries that users pose to web search engines range from two to four terms in length. Much effort has been spent on handling these queries, with considerable success. Apart from advances in retrieval models, simple interaction techniques like query suggestions, spelling correction, and landing page suggestions have also provided substantial improvements in performance. But, what next?
One often-proposed avenue for the next big advance leading to better search experience and user satisfaction is the idea of encouraging users to enter longer queries. A query like 'What are the ideal weather conditions to attend sea kayaking classes in the vicinity of Seattle'? can provide more context for a search engine to work with as opposed to a query like 'sea kayaking Seattle'. However, handling long queries is difficult due to the large number of additional terms that can confuse the search engine. Identifying and appropriately weighting the core concepts in a long query is challenging.
In my talk I will demonstrate the potential for information retrieval using long queries, and show how automatic techniques supported by simple user interaction can lead to 50% improvements in performance on standard test collections. Building on techniques from information retrieval, I show how simple user interaction can be used reformulate long queries, and guide the user effectively towards relevant content. I also address issues such as the efficiency of techniques used to generate the options presented to users and techniques to determine the minimal number of options to present to users. User interaction is a judicious resource that must be carefully invoked: negative experiences can disenchant users. I will also show how we can predict when user interaction for long queries will fail, saving the user time and effort.